RNMarsh said:I'll go test some other low noise transistors.
THx-RNMarsh
Great work, thank you.
Me personally would like to see good old BC327/337 measurements, to confirm Rbb is bellow 30R.
I also wonder if a Fairchild transistor has different noise than an OnSemi transistor? Different manufacturers often have slightly or not so slightly different datasheets for the same part numbers so I'm always wondering if there are some important differences we don't see.
I feel that I really 'stepped in it' responding to the 2219(A) as being a so-so device. It certainly was, 45 years ago, but who knows what they make today?
Still, it is not the best input stage device, no matter, because of low beta at typical input stage operating currents.
After looking at what is easily available today, I hardly know what to say. Back when I was using bipolar transistors, I had a choice of 100's if not 1000's of transistors to optimally choose from, and I did so.
Today, I don't even use bipolar transistors, except for output devices or in my cheapest amps from Parasound. I use almost 100% fets, and have done so for decades, so I don't know what is left on the shelves, especially from quality Japanese parts. My best recommendation is to find an older Toshiba catalog with bipolar transistors, find an interesting pair of devices and then search for them from vendors. They may still have quite a few, and a hobbyist doesn't need 1000's, so this could be practical.
Back in 1980, the Fairchild PE8050-8550, 2A, TO-92 pair had a typical Rbb' of about 5 ohms. This was due to the size of the chip, rather than super engineering. Today, I don't know what they measure today, especially with other brands, that also make them. I found them by accident, from a sample pack from Fairchild, 35 years ago. They were used in the SOTA headamp, made in the early '80's. I could do 0.4nV/rt Hz with a pair.
Still, it is not the best input stage device, no matter, because of low beta at typical input stage operating currents.
After looking at what is easily available today, I hardly know what to say. Back when I was using bipolar transistors, I had a choice of 100's if not 1000's of transistors to optimally choose from, and I did so.
Today, I don't even use bipolar transistors, except for output devices or in my cheapest amps from Parasound. I use almost 100% fets, and have done so for decades, so I don't know what is left on the shelves, especially from quality Japanese parts. My best recommendation is to find an older Toshiba catalog with bipolar transistors, find an interesting pair of devices and then search for them from vendors. They may still have quite a few, and a hobbyist doesn't need 1000's, so this could be practical.
Back in 1980, the Fairchild PE8050-8550, 2A, TO-92 pair had a typical Rbb' of about 5 ohms. This was due to the size of the chip, rather than super engineering. Today, I don't know what they measure today, especially with other brands, that also make them. I found them by accident, from a sample pack from Fairchild, 35 years ago. They were used in the SOTA headamp, made in the early '80's. I could do 0.4nV/rt Hz with a pair.
Thanks for this JC. That would put them in the same class as Hitachi 2sc254x & 2sa108x ie second only to da supa Rohms & Toshibas. 😱Back in 1980, the Fairchild PE8050-8550, 2A, TO-92 pair had a typical Rbb' of about 5 ohms. This was due to the size of the chip, rather than super engineering. Today, I don't know what they measure today, especially with other brands, that also make them. I found them by accident, from a sample pack from Fairchild, 35 years ago. They were used in the SOTA headamp, made in the early '80's. I could do 0.4nV/rt Hz with a pair.
Did you have to screen them for 1/f noise?
I looked at a lot of small medium power devices (and so did Baxandall for QUAD) but the yield was poor. Stopped looking as soon as I found the Hitachis. They had good hfe (and hence good Rni) too.
Last edited:
Not really, these devices were apparently the LARGEST devices made on the low noise, low current line at Fairchild. No problems.
That makes sense.I wonder where the BC3x7 lie?
Noise isn't modeled in transient simulation in LTSpice. Only in noise analysis. You would have to manually add noise voltage sources which is possible, but difficult. However there are BJT model parameters which model the different noise sources that you would see in the noise simulation.
....
THAT transistor models actually have two different versions, one for accurate noise simulation and another for accurate RF simulation. I don't know whether this is necessary or not, it could just be a limitation of the simulator the models were designed for, although the creators apparently payed a lot of attention to model quality.
I'll cautiously suggest that just using the correct value for RB=rbb' (leaving everything else the same) will be good for Audio Noise Analysis .. [*]
Most SPICE models are HF. eg the default LTspice BC307a has RB=rbb'=0r2 😱
Gotta find out where LT sell these supa LN devices 😀
[*] via some simple HF sims but I'm only a SPICE newbie
____________
Bottom line real life
Old "low noise" BJTs like BC560 : RB > 100
Quiet as 2n4401, 2sa970 : RB about 30-40
Supa quiet : RB 5 or less
Last edited:
Although his focus was on RF, Chris Trask confirmed the good performance of the 2N2219/2N2222....you will see the noise level of a 2N2219A TO5 transistor. It is an oldie but goodie. The thing is just under 1nV at 1mA Ic and 8vdc Vc....
Most SPICE models are HF. eg the default LTspice BC307a has RB=rbb'=0r2 😱
After doing some modeling and comparing models to the H-parameter plots on some particularly good datasheets, I really doubt the 0.2R base resistance was used for RF accuracy. I'm serious when I say the vast majority of BJT SPICE models are almost, but not quite, total gibberish. Never ever give the model the benefit of the doubt.
It appears that the Spice models are not always as accurate as they might be.
It doesn't surprise me that professors don't know the subtle differences in available semiconductors. And yet, they are the ones who usually create the Spice models, initially. They usually don't have a QuanTech noise analyzer handy, and they are not heavily sampled by manufacturers, like serious design engineers are.
Today, the withdrawal of the major semiconductor companies from providing superior parts, both jfet and bipolar, makes us attempt to revert to generic devices that were originally made in the '60's. Unfortunately, these are not ideal parts, compared what we could easily get 10 years ago, so we have to be careful in choosing any particular device.
In hi end design and production, we usually have our own store of quality devices that we obtained before they were discontinued by the manufacturer, and we closely monitor new devices from Linear Systems, etc. for future replacement. This is how Ayre, Constellation, Nelson Pass, and other hi end companies make SOTA designs, today. Sorry that I can't be more helpful.
It doesn't surprise me that professors don't know the subtle differences in available semiconductors. And yet, they are the ones who usually create the Spice models, initially. They usually don't have a QuanTech noise analyzer handy, and they are not heavily sampled by manufacturers, like serious design engineers are.
Today, the withdrawal of the major semiconductor companies from providing superior parts, both jfet and bipolar, makes us attempt to revert to generic devices that were originally made in the '60's. Unfortunately, these are not ideal parts, compared what we could easily get 10 years ago, so we have to be careful in choosing any particular device.
In hi end design and production, we usually have our own store of quality devices that we obtained before they were discontinued by the manufacturer, and we closely monitor new devices from Linear Systems, etc. for future replacement. This is how Ayre, Constellation, Nelson Pass, and other hi end companies make SOTA designs, today. Sorry that I can't be more helpful.
I think that it would be useful to discuss what the 'important' parameters of a bipolar transistor are.
You know: Base resistivity
Voltage breakdown, F(t), but most importantly: BETA (with current) How much and how linear.
Finally, power dissipation.
You know: Base resistivity
Voltage breakdown, F(t), but most importantly: BETA (with current) How much and how linear.
Finally, power dissipation.
for line level and low level and power levels...... for each--- what are the trade-offs and which becomes more important for these three apps.
Then, which parameters are accurate from spec sheets and what usually needs to be measured. There are a few circuits developed for DIY'ers for measuring very low noise levels.... might be applied/adapted to transistor noise testing so all can have their own transistor noise analyzer?
THx-RNMarsh
Then, which parameters are accurate from spec sheets and what usually needs to be measured. There are a few circuits developed for DIY'ers for measuring very low noise levels.... might be applied/adapted to transistor noise testing so all can have their own transistor noise analyzer?
THx-RNMarsh
Last edited:
HD analog is in the details -
In screwing around with noise measurements.... the nV noise changes with current and voltage... of course....sometimes quite a lot -- Just changes in signal level is going to change the noise dynamically. Maybe most dramatic change is from high dynamic range voltage/current swing circuits [line stage and power amps]. Might not notice it as noise [in steady state sense].... but how does that translate into clarity of musical notes etc?
More reasons to run in an optimized rich class A mode and using low noise transistors even in power amps (cascoded, most likely).
THx-RNMarsh
In screwing around with noise measurements.... the nV noise changes with current and voltage... of course....sometimes quite a lot -- Just changes in signal level is going to change the noise dynamically. Maybe most dramatic change is from high dynamic range voltage/current swing circuits [line stage and power amps]. Might not notice it as noise [in steady state sense].... but how does that translate into clarity of musical notes etc?
More reasons to run in an optimized rich class A mode and using low noise transistors even in power amps (cascoded, most likely).
THx-RNMarsh
Last edited:
Thanks for this.Although his focus was on RF, Chris Trask confirmed the good performance of the 2N2219/2N2222.
LN at HF is probably the most demanding of good SPICE models as stuff like rbb' has to be correct for both HF performance and well as noise.
Alas, many devices which are LN at HF are poor at audio cos 1/f rise in noise.
Many FETs have this sin but there are exceptions like J305 which, though it does not have 1nV/rtHz like 2sk170, actually gives less noise in real life capacitor mikes.
And of course there are BJT exceptions to this like 2n4401/3 too.
The nice thing about LN HF datasheets is they usually shows performance at high currents (usually for Zs 50R) so is a good indicator for low rbb'. The caveat, as above, is 1/f noise.
LN is a lot about clean processing. Today, only some makers dare to put their reputations and noise specs on their datasheets.
Only Fairchild seems to be in this august group now, and their new stuff appears consistently good ... especially for JFETs which are my prime practical interest this Millenium
Mr. Marsh, in a 'perfect' BJT, Env is inversely proportional to collector current so Yes. It will change ALOT.RNMarsh said:the nV noise changes with current and voltage... of course....sometimes quite a lot -
This is what JC was referring to in #3100.
Looking to see how how a real device departs from this behaviour gives us rbb' and other stuff that helps us to design good LN amps.
Last edited:
Remember guys that the voltage noise of a bipolar transistor is equal 1/2Gm+Rbb'. At 1ma this is 13+Rbb'; at 10ma is 1.3+Rbb' so usually we don't operate that high in current, except for very low Rbb' devices.
Of course, at 100uA the effective noise RESISTANCE is 130+Rbb', so now BETA becomes more important, because we usually have to figure in an I(noise) contribution that is proportional to beta with source resistances over 50 ohms or so. This is where the beta droop of the 2N2219 could really effect things, and another, higher beta (at 100ua) device might be better, even if the Rbb' is somewhat higher.
Of course, at 100uA the effective noise RESISTANCE is 130+Rbb', so now BETA becomes more important, because we usually have to figure in an I(noise) contribution that is proportional to beta with source resistances over 50 ohms or so. This is where the beta droop of the 2N2219 could really effect things, and another, higher beta (at 100ua) device might be better, even if the Rbb' is somewhat higher.
BC377 measured .6nV
Hi,
I just put on a BJT type BC377 (NPN) and it lives up to mfr spec....
at 1Khz/10KHz/100KHz the noise per sqrt Hz is .6nv/.65nV/.53nV
At 100Hz it is 3.5nV
All at 3mA/15vdc; beta = 700
THx-RNMarsh
Hi,
I just put on a BJT type BC377 (NPN) and it lives up to mfr spec....
at 1Khz/10KHz/100KHz the noise per sqrt Hz is .6nv/.65nV/.53nV
At 100Hz it is 3.5nV
All at 3mA/15vdc; beta = 700
THx-RNMarsh
Last edited:
Mr. Marsh, in a 'perfect' BJT, Env is inversely proportional to collector current so Yes. It will change ALOT.
This is what JC was referring to in #3100.
Looking to see how how a real device departs from this behaviour gives us rbb' and other stuff that helps us to design good LN amps.
I havent found any perfect transistor behaviour, yet.
THx-RNMarsh
Although his focus was on RF, Chris Trask confirmed the good performance of the 2N2219/2N2222.
FWIW, a quad (4x) 2N2222 is available in a SOIC-16 package, the MMPQ2222. I have no idea how closely they're matched, but in principle it should be possible to build a low-noise LTP with two paralleled NPNs on each side of the LTP, for a total of 4 NPNs, all in the same package, which is nice for thermal inertia.
FWIW, a quad (4x) 2N2222 is available in a SOIC-16 package, the MMPQ2222. I have no idea how closely they're matched, but in principle it should be possible to build a low-noise LTP with two paralleled NPNs on each side of the LTP, for a total of 4 NPNs, all in the same package, which is nice for thermal inertia.
Hah! I just happened to have one on my desk. Several of these types are still around with complimentary pairs also. All in DIP [originally Motorola Plastic Quads]

THx-RNMarsh
Last edited:
All in DIP [originally Motorola Plastic Quads]
Funky - that's a DIP-14 (which is enough for 4 isolated transistors), but the SMD MMPQ version is SOIC-16, IIRC. For small discrete designs, a TSOP or QFP-12 version might be more useful. Actually, a matched dual in SOT363 or SOT457 would do just fine.
Last edited:
With a quad you need ROM for 4 dies, which may be pretty large. There are good dual bipolars still available but fewer. If you need GASfets or other exotic rf parts there is new stuff. That's where the money is. In the low frequency region you find fast power FETs and integrated stuff, again where the money is. Discrete bipolar is trailing tech and no exec will commit $$ to that market.
- Home
- Source & Line
- Analog Line Level
- Discrete Opamp Open Design